AI Insights & Deep Dives

Featured

Engineering 9 min read

External Agent Skills Worth Trying, and the Filter We Run Them Through

New agent skills land faster than anyone can evaluate them, and plenty are a paragraph of vibes in a Markdown file. Here are the external ones our team actually kept, why they survived, and the four questions we ask before installing anything.

D Duy (Đỗ Anh)

Aug 24, 2026 Read →

Featured

Engineering 12 min read

Convert Everything to Markdown for AI Agents — Part 4: anydoc vs markitdown vs /utils:pdf

Three converters, six formats, then 55 real PDFs to check how often the failures actually happen. anydoc is 30× faster and silently drops ligatures in LaTeX documents; markitdown produced one heading in the whole corpus. Here is the full benchmark and the routing rule we ended up with.

D Duy (Đỗ Anh)

Aug 24, 2026 Read →

Featured

AI 13 min read

Convert Everything to Markdown for AI Agents — Part 3: Word Documents

DOCX is OOXML, the same family as PowerPoint, so it should finally be the easy entry in this series. It mostly is — except for a handful of sharp bugs in Word's own formatting model, and one bigger problem that only shows up once you try to run the conversion on a schedule instead of by hand.

T Trobz

Jul 31, 2026 Read →

Featured

AI 8 min read

Convert Everything to Markdown for AI Agents — Part 2: PowerPoint Decks

PPTX is structured XML, so it should be easy to read — and it still isn't. A deck's meaning is spatial: split across slides, hidden in speaker notes, locked in charts and image-only slides. Here's how to convert one to clean Markdown without dropping it.

T Trobz

Jun 18, 2026 Read →

Featured

AI 5 min read

How Claude Cowork Queries Your Odoo Instance Using the Trobz Odooly Skill

Claude Code becomes a lot more useful when it can read your actual Odoo data. Five minutes of setup — install the public skills plugin, configure odooly.ini — and your agent goes from guessing to querying.

T Thảo (Đoàn Thị)

Jun 11, 2026 Read →

Featured

Engineering 8 min read

What Agent Skills Are, and Why Our Team Builds Them

Your best Odoo developer has a prompt they retype every morning, and it lives nowhere but their head. A skill is how you take that knowledge out of the chat box and make it reusable.

D Duy (Đỗ Anh)

May 25, 2026 Read →

Featured

Engineering 6 min read

Guardrails in AI Chatbot: Protecting the System from Input to Output

Security needs to be part of RAG chatbot design from day one. Here's how guardrails work, where to place them, and how to handle them without leaking information to attackers.

T Thiên (Võ Hồng)

May 21, 2026 Read →

Featured

Engineering 9 min read

Agentic RAG in Practice: From Single-Pass Retrieval to a Self-Correcting Pipeline

Basic RAG breaks in predictable ways. Here's how we replaced a pile of conditional patches with an agentic graph pipeline, and what it actually cost in latency, complexity, and accuracy.

T Thiên (Võ Hồng)

May 21, 2026 Read →

Strategy 7 min read

The Most Valuable PoC We've Delivered Said "Don't Build This"

A distribution company hired us to automate purchase order approvals with AI. Two weeks later, we told them not to. Here's why that was the right outcome — and what we built instead.

T Trobz

May 15, 2026 Read →

AI 9 min read

Hybrid Search in Odoo: Combining BM25 and Vector Similarity for Better Retrieval

Pure vector search misses exact keyword matches; pure BM25 misses semantic intent. Hybrid search combines both — here's how to implement it on Odoo data with pgvector and PostgreSQL full-text search.

T Trobz

May 14, 2026 Read →

Strategy 8 min read

The PoC-to-Production Gap: What Changes When Real Users Touch Your AI

A PoC that works on your laptop is not a production system. The gap between the two is wider than most stakeholders expect, and it shows up in six predictable places.

T Trobz

May 13, 2026 Read →

AI 9 min read

Vietnamese Vendor Contracts: The Corner Cases That Break Every OCR Pipeline

After processing 3,000+ Vietnamese B2B contracts, the clean PDFs were never the problem. Here's what actually broke our OCR pipeline — and what we did about it.

T Trobz

May 12, 2026 Read →

AI 10 min read

Building a RAG-Powered Odoo Support Agent That Knows Your Product Catalogue

A step-by-step guide to building an Odoo support agent that retrieves product data, pricing rules, and quote history to answer sales rep questions accurately — and knows when to say it doesn't know.

T Trobz

May 11, 2026 Read →

Strategy 8 min read

Scoping a PoC: The One-Page Brief That Gets Stakeholders Aligned Before You Write a Line of Code

Most AI PoCs fail not because the model was bad, but because nobody agreed upfront what 'success' meant. One page fixes that.

T Trobz

May 10, 2026 Read →

AI 7 min read

When the Model Was Right and the Business Was Wrong: A Corner Case Retrospective

Our demand forecasting model correctly flagged a stockout risk three weeks out. It also generated 40 false positives — 12 of them caused by a data entry error the model had silently learned. Here's how we ran the retrospective.

T Trobz

May 9, 2026 Read →

AI 8 min read

Semantic Search over Odoo Knowledge Base Articles: From Zero to Production in an Afternoon

Odoo's built-in full-text search misses conceptually related content. Here's how to build semantic search over knowledge.article using pgvector and sentence-transformers — end to end, with benchmarks.

T Trobz

May 8, 2026 Read →

Strategy 8 min read

Evaluating RAG Quality: Beyond "Does It Answer the Question?"

Most teams evaluate RAG by asking: does it answer correctly? That question catches less than half the failures. Here's the full evaluation stack.

T Trobz

May 7, 2026 Read →

Featured

AI 7 min read

Convert Everything to Markdown for AI Agents — Part 1: PDF Files

Agents don't read PDFs. They read text. Here's what happens when you feed them one, and how to convert PDFs to Markdown without silently losing structure.

T Trobz

May 7, 2026 Read →

Automation 7 min read

How a 2-Week PoC Changed a 6-Month Project Estimate

A procurement automation project initially scoped at 6 months was reframed in 2 weeks. Real Odoo data, real accuracy numbers, and a conversation that changed the moment there was something to measure.

T Trobz

May 6, 2026 Read →

Featured

Engineering 8 min read

Tips and Tricks to Lower Your Claude Code Token Consumption

In a long Claude Code session, most tokens go to re-reading conversation history, not generating output. Here is how to measure and fix that.

D Duy (Đỗ Anh)

May 5, 2026 Read →

Strategy 8 min read

From Slide Deck to Running System in 14 Days: Inside a Trobz AI PoC Sprint

Most AI consultancies deliver a slide deck and call it discovery. We deliver a running system. Here's what happens in the 14 days between kickoff and go/no-go.

T Troboz

Apr 30, 2026 Read →

AI 9 min read

Product Recommendation Engine for Odoo eCommerce: Implementation Guide

Collaborative filtering on your existing sale.order.line data is enough to build a working product recommendation engine — no external ML platform, no separate vector database. Here's how to implement it end-to-end inside Odoo eCommerce.

T Trobz

Apr 29, 2026 Read →

Strategy 7 min read

Human-in-the-Loop Is Not a Limitation -- It's the Whole Point

Autonomous agents that can't be overridden won't be trusted -- and agents that aren't trusted won't be used. Human-in-the-loop isn't a design compromise; it's the feature that makes deployment possible.

T Trobz

Apr 28, 2026 Read →

Automation 8 min read

How a Manufacturing Company Automated Delivery Exception Alerts Across 3 Warehouses

When three warehouses feed one Slack channel, every exception looks equally urgent. Here's how we separated the noise from the fires — and cut critical response time from 4 hours to 18 minutes.

T Trobz

Apr 27, 2026 Read →

AI 9 min read

Cash Flow Forecasting in Odoo: A Practical Guide to the ML Pipeline

Your Odoo database already has everything needed for a 6-week cash flow forecast. Here's how to build the ML pipeline from raw data to a live dashboard widget.

T Trobz

Apr 26, 2026 Read →

Strategy 7 min read

Composable ERP in Practice: What Gartner Got Right and What They Left Out

Gartner's composable ERP model points in the right direction. But after implementing it on top of Odoo, the gaps in the theory become obvious fast.

T Trobz

Apr 25, 2026 Read →

AI 7 min read

AI-Powered Anomaly Detection Caught a $40K Billing Error in Week One

A finance team deployed anomaly detection on their Odoo Accounting data. In the first week, it flagged a duplicate vendor bill for $40,000 that had already passed manual review twice.

T Trobz

Apr 24, 2026 Read →

Featured

Events 4 min read

Trobz at the Alibaba Cloud AI Growth Workshop

On April 23, the Trobz team attended the Alibaba Cloud AI Growth Workshop in Ho Chi Minh City: a morning of sharp insights on cloud infrastructure, Qwen AI models, and what Generative AI looks like at global scale.

T Thiên (Võ Hồng)

Apr 23, 2026 Read →

Featured

AI 8 min read

Em Dashes Are the New AI Tell. Replace-All Is Not the Fix.

Em dashes have become the most-cited tell of AI writing. Search-and-replace is the wrong fix — the problem is the sentence structure underneath them.

T Trobz

Apr 23, 2026 Read →

AI 8 min read

Building an OCR Pipeline for Bills of Lading That Feeds Odoo Inventory

Bills of lading arrive in dozens of formats, but your Odoo Inventory needs clean, structured data. Here's how to build a pipeline that goes from PDF scan to stock.picking automatically.

T Trobz

Apr 23, 2026 Read →

Featured

Engineering 13 min read

Finding the right company photo: how we indexed thousands of Trobz images with AI

Trobz has accumulated over 8,000 photos across 15 years of events, trips, and photoshoots. We built an AI-powered search tool to find and rank the best ones — so we can stop relying on stock photos for our own blog.

D Duy (Đỗ Anh)

Apr 22, 2026 Read →

Strategy 7 min read

AI Roadmapping for Odoo Companies: Start with the Bottleneck, Not the Buzzword

Most AI roadmaps start with the technology. The ones that actually ship start with a specific process that's costing someone real hours every week.

T Trobz

Apr 22, 2026 Read →

AI 8 min read

The Overdue Invoice Chaser Agent: From Configuration to First Collection

Most overdue invoice processes fail not because the team is lazy but because the workflow is inconsistent. Here's how an AI agent changes that — and what the first month looks like.

T Trobz

Apr 21, 2026 Read →

Featured

Engineering 4 min read

Enterprise RAG Strategy: The Foundational Architecture for Business AI

RAG bridges the gap between LLMs and private enterprise knowledge. This article breaks down the architecture, design factors, trade-offs, and ROI decision matrix for building RAG into your business AI stack.

T Thiên (Võ Hồng)

Apr 20, 2026 Read →

AI 9 min read

Adding Semantic Search to Your Odoo Product Catalogue with field_vector

Keyword search breaks down when your sales reps don't know exact product names. Here's how to add semantic natural-language search to your Odoo product catalogue using the field_vector OCA module — no extra database required.

T Trobz

Apr 20, 2026 Read →

Strategy 8 min read

Open Source AI for Enterprise: Why We Chose n8n, pgvector, and Transformers.js

Every AI vendor promises control and auditability. We chose open-source tools that actually deliver them — n8n, pgvector, and Transformers.js — and here's the reasoning behind each decision.

T Trobz

Apr 19, 2026 Read →

AI 7 min read

Late Payment Predictor: How an Odoo-Based Manufacturer Reduced DSO by 11 Days

An ML model trained on Odoo payment history predicted late payers with 79% precision — and cut Days Sales Outstanding by 11 days in four months.

T Trobz

Apr 18, 2026 Read →

Automation 10 min read

Building a Low-Stock Alert System with n8n and Odoo Inventory in Under an Hour

Poll Odoo's stock.quant model via JSON-RPC, apply per-product reorder rules, group alerts by warehouse, and deliver one consolidated Slack or email notification — no Odoo customization required.

T Trobz

Apr 17, 2026 Read →

Strategy 8 min read

Data Integrity Is the Moat: Why AI Can't Shortcut 20 Years of ERP Schema Design

Foreign key constraints, sequence locking, fiscal year gates, and multi-company isolation are not just implementation details. They represent 20 years of encoded business rules. An AI-generated schema misses exactly the constraints that matter most.

T Trobz

Apr 16, 2026 Read →

Featured

Engineering 8 min read

The RPI Workflow: Context Engineering for Odoo Module Development

Claude Code goes sideways once its context window fills past 40%. Research, Plan, Implement: three sessions, three files, keeps it on track.

T Trí (Đoàn Minh)

Apr 15, 2026 Read →

AI 7 min read

Processing Expense Receipts at Scale: Document AI for a 200-Person Services Company

A 200-person services firm was spending 3,000+ staff-hours a year manually keying expense receipts into Odoo. Here's the Document AI pipeline that cut that to near zero.

T Trobz

Apr 15, 2026 Read →

AI 11 min read

Autonomous Bank Reconciliation in Odoo: How the Agent Decides What to Match

Bank reconciliation is one of the highest-ROI targets for an Odoo AI agent — repetitive, high-volume, but nuanced enough that the matching logic matters. Here's exactly how we built it.

T Trobz

Apr 14, 2026 Read →

Strategy 7 min read

The Hidden Cost of AI Prototypes That Never Ship

Most enterprise AI PoCs are built to impress a room, not to run in production. The gap between a working demo and a deployed system involves monitoring, retraining pipelines, error handling, and ownership — none of which appear in a two-week sprint.

T Trobz

Apr 13, 2026 Read →

AI 8 min read

Smart Lead Scoring in Practice: How a SaaS Sales Team Stopped Guessing

A SaaS sales team was spending 60% of its time on leads that never converted. AI lead scoring built directly into Odoo CRM changed that — without a data warehouse or a dedicated data team.

T Trobz

Apr 12, 2026 Read →

AI 9 min read

Training a Churn Predictor on Odoo CRM Data Without a Data Warehouse

Your Odoo database already contains the signals that predict customer churn. This tutorial shows how to extract them, train a classifier, and surface risk scores directly in the CRM kanban view — no data warehouse required.

T Trobz

Apr 11, 2026 Read →

Strategy 7 min read

When to Say No to an AI Project: Five Patterns We've Learned to Walk Away From

Most AI projects don't fail because of bad models or wrong tools. They fail because they should never have started. Here are five patterns that tell us when to walk away.

T Trobz

Apr 10, 2026 Read →

Automation 8 min read

From 47 Emails a Day to One Dashboard: Automating Purchase Order Approvals in Odoo

A procurement team was drowning in email-based PO approvals — 47 a day, scattered across inboxes, with no audit trail. Here's how n8n and Odoo fixed it.

T Trobz

Apr 9, 2026 Read →

Featured

AI 7 min read

MCP Web Search for AI Agents: Tavily vs Brave Search

I tested two popular MCP web search servers, Tavily and Brave Search, while building AI agents with Claude. Here's what I found, with real benchmark numbers and a clear recommendation.

T Thảo (Đoàn Thị)

Apr 7, 2026 Read →

Featured

Engineering 7 min read

Connecting AI Agents to Odoo with MCP: A Practical Overview

Several MCP implementations exist for connecting AI assistants to Odoo. The tooling is improving fast. Authentication is the hardest part to get right.

T Thiên (Võ Hồng)

Apr 6, 2026 Read →

Strategy 7 min read

Why "AI-Native ERP" Is a Marketing Term, Not an Architecture

Vendors claim their ERP is "AI-native." The phrase means nothing precise and obscures the question that actually matters: does the integration layer let you connect AI capabilities as they evolve?

T Trobz

Apr 4, 2026 Read →

AI 8 min read

How a Hanoi Distributor Cut Invoice Processing from 4 Hours to 12 Minutes

A Hanoi-based distributor was spending 4 staff-hours a day keying in vendor invoices. A Document AI pipeline brought that to 12 minutes. Here is what it took.

T Trobz

Apr 3, 2026 Read →

Featured

Engineering 5 min read

AI in Odoo: Two Architectures, Two Different Bets

Odoo SA and the OCA take fundamentally different approaches to integrating AI. One bundles AI into the platform. The other keeps it deliberately outside. Here is what that looks like in practice.

T Thiên (Võ Hồng)

Apr 2, 2026 Read →

Automation 9 min read

5 n8n Workflows Every Odoo Finance Team Should Steal

Five production-ready n8n workflows for Odoo Finance: invoice reminders, reconciliation alerts, budget notifications, expense routing, and month-end triggers — all built on the same JSON-RPC pattern.

T Trobz

Apr 2, 2026 Read →

Strategy 7 min read

The AI Readiness Audit: 7 Questions to Ask Before You Touch Your ERP

Most AI projects in Odoo fail before a single model runs. The problems are always upstream: dirty data, undefined processes, nobody who owns the outcome. Here's how to find them first.

T Trobz

Apr 1, 2026 Read →

Featured

Engineering 11 min read

Agentic AI for QA: 3 different approaches

AI agents can now replace the manual post-deploy verification step. But which agent, and at what level of access? We ran three against the same Odoo 18 ticket to find out.

T Thiên (Võ Hồng)

Apr 1, 2026 Read →

Strategy 6 min read

Is the AI Revolution Just the Excel Revolution All Over Again?

Vibe coding feels powerful, until you inherit someone else's 800-line VBA macro. The AI moment rhymes with the spreadsheet moment in ways that should make every engineering team pay attention.

T Trobz

Mar 28, 2026 Read →

Featured

Automation 9 min read

How to Alert Uninvoiced Sales Orders in Odoo Using n8n (Without Creating Email Noise)

n8n + Odoo 18 automation that flags confirmed but uninvoiced sales orders: one consolidated email every 15 minutes, not one per order.

T Thái (Hồ Hoàng)

Mar 28, 2026 Read →

Strategy 6 min read

Composable ERP: The Architecture Shift That Makes AI Integration Actually Work

Traditional monolithic ERP systems were not designed to absorb AI capabilities incrementally. Composable ERP, built from interchangeable API-connected modules, is the architecture that makes it possible.

T Trobz

Mar 28, 2026 Read →

Strategy 6 min read

Can an ERP Built with Claude Replace Odoo?

Someone always builds the thing you said couldn't be built. But there's a precise difference between an impressive invoicing app and an ERP, and it matters more than most people realise.

T Trobz

Mar 28, 2026 Read →

Featured

AI 5 min read

Transformers in the browser? A concrete example: semantic search over 3,000+ Odoo modules

How we built a fully client-side semantic search engine for 3,000+ Odoo modules with no backend, no API key, no internet required after first load.

D Duy (Đỗ Anh)

Mar 1, 2026 Read →

Featured

AI 2 min read

Introduction to the field_vector OCA module for Odoo

Discover how the field_vector OCA module brings native vector field support to Odoo, enabling semantic search and RAG without a separate vector database.

H Hải (Lê Nguyễn)

Feb 10, 2026 Read →